-
Notifications
You must be signed in to change notification settings - Fork 8
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[WIP] Rework of allocations #78
Conversation
Would it then maybe be reasonable to think of a new way to do the allocation? The main challenge I think is – in the example of the log – to allocate a tangent vector when no tangent vectors are among the input parameters? I have not yet given it so much of thoughts, but in general found the current approach ok to work with. |
That's one challenge. There is also the problem of embedded manifolds where we have to differentiate between ambient space representation and embedded manifold representation (where we have Of course, the current approach doesn't solve all of these problems but they are relatively niche and it's not that hard to patch an issue when it's actually encountered. I wanted to see how complicated this rework would be so I tried it but it doesn't seem to be worth the trouble. |
Ok, is there still a doable way to reduce the current ambiguities? |
Sort of yes, it just requires solving them one by one: writing a new method, and then making a test which is sometimes not straightforward. |
Is this still something we should follow up on? Or is this obsolete after 3 years? |
I think this PR is not worth continuing because other PRs have made some smaller improvements. The current system is not perfect but its deficiencies show up so rarely that I'm not very motivated towards improving it. |
I have no clue where the imperfections lie, but if a discussion help, we can have one about that for sure. I feel it works very well overall. |
The main deficiency is in the number of |
This is my semi-broken attempt at improving out allocation methods. I'm not sure if this is worth pursuing -- very little is simplified and other places get more complicated. It also shows how most currently existing methods of
allocate_result
are poorly design in some way but fixing it is a really tedious task.